ADVANCED MATERIALS & PROCESSES | JULY/AUGUST 2025 12 MACHINE LEARNING | AI AI HELPS OPTIMIZE SOLID-STATE BATTERIES Researchers at Tohoku University, Japan, developed a data-driven AI framework that suggests promising solid-state electrolyte (SSE) candidates for sustainable energy applications including solid-state batteries. The model not only selects optimal candidates but also predicts how the reaction will occur and why the candidate is a good choice, helping scientists make signifi- cant headway before running any physical experiments. “The model essentially does all of the trial-and-error busywork for us,” says Professor Hao Li. “It draws from a large database from previous studies to search through all the potential options and find the best SSE candidate.” The new AI framework integrates large language models, MetaD, multiple linear regression, genetic algorithm, and theory-experiment benchmarking analysis. The predictive models draw from both experimental and computational data. A goal of this study was to understand the structure-performance relationships of SSEs. The model predicts activation energy, identifies stable crystal structures, and improves overall workflow. The findings demonstrate that MetaD is an optimal computational technique that shows high levels of agreement with experimental data for complex hydride SSEs. Further, by combining feature analysis with multiple linear regression, they constructed precise predictive models for the rapid evaluation of hydride SSE performance. The researchers believe their work will help enable more efficient design of next-generation solid-state batteries. www.tohoku.ac.jp. MACHINE LEARNING PREDICTS MATERIAL FAILURE Two scientists at Lehigh University, Bethlehem, Pa., predicted abnormal grain growth in simulated polycrystalline materials for the first time. They say the development could lead to better materials for high-stress environments such as combustion engines. So far, predicting abnormal grain growth has been extremely challenging due to the numerous combinations that can go into making any given alloy. The new computational simulation helps narrow down possibilities by eliminating Cross-sections shown every 10M MCS (Monte Carlo steps). The highlighted (red) grain becomes abnormally large just after 67M MCS. The Lehigh team predicted this grain would become abnormal using only the data from 11 to 15M MCS, long before abnormality occurs. materials that are likely to develop abnormal grain growth. The challenge is that abnormal grain growth is a rare event and early on, grains that will become abnormal look just like the others. To address this, the team developed a deep learning model that combines two techniques to analyze how grains evolve over time and interact with each other: A long short-term memory network models how the material properties would be evaluated while a graph-based convolutional network establishes relationships between the data that could then be used for prediction. Critical to early detection was using the models to examine a grain’s characteristics over time before the abnormality occurred. The team aligned each simulation at the point in time where the grain became abnormal and then worked backward examining its evolving properties. By identifying consistent trends in these properties, they were able to predict which grains would become abnormal. The goal, says researcher Brian Chen, is to identify materials that are highly stable and can maintain their physical properties under a wide range of high-temperature, high-stress conditions. lehigh.edu. A new AI framework suggests promising solid-state electrolyte candidates to build better batteries.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==